Abstract: Abstract: Using unmanned aerial vehicle (UAV) remote sensing monitoring system can rapidly and cost-effectively provide crop physiological traits for crop breeding. UAV equipped with an imaging spectrometer to estimate soybean yield is of great significance for high-throughput and rapid access to large-scale soybean production. However, different sampling areas led to different spectral data, thus affecting the accuracy of soybean grain yield. The objective of this study was to explore the influence of different sampling area on the measuring accuracy of soybean yield, and to analyze the optimum sampling area for estimating soybean grain yield. A 3-by-275 field experiment was performed in 2015, which was arranged in a randomized complete block design with 3 repetitions. An agricultural UAV remote sensing monitoring system was established by a multi-rotor UAV equipped with Cuber UHD185 Firefly imaging spectrometer (Cubert UHD185). Based on this system, the UAV flight experiments were conducted in Jiaxiang County, Shandong Province at multifarious reproductive growth stages, including the period from the initial blossoming stage to the fully blossoming stage (R1-R2), the initial pod stage (R3), from the full pod stage to the initial seed stage (R4-R5), the full seed stage (R6) and from the full seed stage to the mature stage (R6-R7). In order to get stable soybean canopy hyperspectral data, the calm and cloudless weather was selected to conduct the experiment. Hyperspectral data of each block were obtained according to the vector image georeferenced with the hyperspectral image. Since soybean yield was highly correlated with canopy reflectance measured by the UAV with Cubert UHD185 system in R4-R5 stages, the hyperspectral data obtained in R4-R5 stages were used to be further analyzed. Firstly, softwares such as Cubert-Pilot from Cubert Company and Agisoft PhotoScan from Agisoft LLC Company were used to realize image mosaic. The length and width of every block were minified in equal proportion for 20 times, and thus 21 sampling areas were gained, which were then used as vector images to get 21 groups of hyperspectral data. Next, 4 vegetation indices, i.e. the green normalized difference vegetation index (GNDVI), the normalized difference vegetation index (NDVI), the ratio vegetation index (RVI) and the modified soil-adjusted vegetation index-2 (MSAVI2), were calculated from the spectral information extracted from 21 different sampling areas. Thirdly, analysis of variance (ANOVA) was performed, and the result revealed that the selected canopy vegetation index was closely related to the growing conditions of different soybean varieties. After that, the partial least squares regression (PLSR) models were developed to predict the yield using the 4 vegetation indices obtained from 21 different sampling areas, with the r value up to 0.8117 (the number of sample points for modeling was 270, P<0.01). And the best sampling area was further confirmed to 9.03-10.13 m2 according to the changing trend of correlation coefficients. Namely, when the ratio of length and width of the sampling area to that of the total block was between 4.25:5 and 4.5:5, the obtained canopy spectra could estimate the soybean yield as accurately as possible. The study confirmed that using the UAV with Cubert UHD185 for screening and predicting soybean yield was practical, with the R2 up to 0.659. The method used in this study to select the optimum sampling area and the result of this study according to the optimum spatial sampling are expected to provide technical support for the analysis of the crop phenotype information using high or low altitude hyperspectral images.